New Framework for Statistical Testing on Directed Graphs Using Surrogate Data
Researchers extend surrogate generation to directed graphs, enabling robust hypothesis testing.
Deep Dive
Researchers propose a new framework for statistical hypothesis testing on directed graphs using surrogate data generation. They first define wide-sense stationary signals on directed graphs via eigendecomposition of the graph shift operator. Then they generate surrogate signals preserving covariance structure under stationarity, enabling construction of null distributions. Real data tests show superiority over existing undirected or permutation-based methods, according to the article.
Key Points
- Defines wide-sense stationary signals on directed graphs using eigendecomposition of the graph shift operator.
- Generates surrogate graph signals that preserve covariance structure under stationarity assumptions.
- Outperforms naive permutation and undirected surrogate methods on real directed network data (e.g., brain connectivity).
Why It Matters
Enables statistically rigorous hypothesis testing on directed networks like brain connectivity or citation graphs.